Cross-Lingual Reusability of LLM Pipelines: Financial Extraction, Multilingual Reasoning, and Semantic Screening

A study posted to arXiv (2610.27043v1) found that English-built LLM pipelines can be reused across languages with qualifications: Gemini 3.5 Flash reached 95.2–97.6% cheque-extraction accuracy across prompt languages, and FinanceBench per-language means spanned 5.8 percentage points (77.1–82.9%) over 50 items from SEC 10-K filings in four languages. Higher reasoning effort produced only marginal gains on general-knowledge QA but up to 8 percentage points on financial content, while citation-set overlap with English averaged 77–79% Jaccard across the nine languages tested on Mintaka. The authors conclude that per-domain, per-language validation remains necessary because neither semantic similarity nor self-reported confidence reliably predicts cross-lingual accuracy.

arXiv:2610.27043v1 Announce Type: new Abstract: Financial institutions increasingly deploy large language models LLMs in multilingual markets, raising a practical question: can pipelines developed and validated in English be reused across languages, or must they be rebuilt per locale? We test cross-lingual reusability along three dimensions--accuracy, reasoning-effort transfer, and behavioural stability--through three complementary tasks. Task 1 cheque extraction and Task 2 Mintaka QA with pinned evidence across nine languages and nine model-effort configurations test whether accuracy and effort settings transfer; Task 3 FinanceBench pilot, 50 items over SEC 10-K filings in four languages validates whether these patterns generalise to financial content. Accuracy reusability is broadly confirmed: Gemini 3.5 Flash achieves 95.2-97.6% extraction accuracy across prompt languages, and FinanceBench per-language means span 5.8pp 77.1-82.9% , though the language-difficulty ranking reverses between domains Hindi is worst on Mintaka, best on FinanceBench at 82.9% . Effort reusability is domain-dependent: higher reasoning effort yields marginal gains on general-knowledge QA but up to 8pp on financial content, and effort configurations that help in English help proportionally in other languages. Behavioural reusability holds: citation-set overlap with English averages 77-79% Jaccard, with no single language consistently worse. Together, these results support qualified reuse of English-built pipelines, but per-domain, per-language validation remains necessary--neither semantic similarity nor self-reported confidence reliably predicts cross-lingual accuracy.